Previously, we proposed the ‘oscillometric finger pressing method’ for BP monitoring via a smartphone¹⁹. The user presses their fingertip against the smartphone held at heart level to steadily increase the external pressure of the underlying artery, while a custom PPG-force sensor unit attached to the phone measures the resulting variable amplitude blood volume oscillations and applied finger pressure. The phone also visually guides the amount of finger pressure that the user applies over time and computes systolic and diastolic BP from the constructed oscillogram. We later showed that this method could be implemented as an iPhone X app that leverages the front camera to measure the PPG waveform and the 3D touch sensor under the adjacent screen to measure finger force¹⁷. Although exquisite, the 3D touch sensor was not designed for BP measurement and was thus suboptimal (e.g., the finger force could saturate prematurely for higher BP or larger fingers). Moreover, the finger contact area could only be estimated, which introduced significant error in the BP measurement. Apple discontinued 3D touch in subsequent versions of the iPhone (11 and onwards) anyhow, and few, if any, phones today include fine force sensing.

The smartphone app that we have introduced herein circumvents the need for a force sensor by varying the internal BP via hand raising and accurately measuring the hydrostatic pressure change with the accelerometer. In conventional oscillometry, the external pressure of the artery is swept. The clever idea of implementing oscillometry via internal BP changes induced by hand raising was proposed by Shaltis et al.^20,21,22. These investigators built a finger worn ring comprising a PPG sensor, accelerometer, and force sensor. This ring could potentially measure systolic and diastolic BP from the function relating the blood volume oscillation amplitude to the difference between the contact pressure of the ring on the finger and the hydrostatic pressure change (ρgh, via the accelerometer). However, the investigators did not address how to attain and maintain a finger contact pressure around the mean BP to obtain a full oscillogram. This step is not only an enabling one but also nontrivial, because mean BP is essentially what is sought for measurement. Furthermore, they only published results from five volunteers, so the ‘oscillometric hand raising method’ has remained mostly theoretical until now.

Our overall idea was to translate the oscillometric hand raising method for implementation as a smartphone app (see Fig. 1). We needed to conceive two key innovations to realize the app. Firstly, we devised a procedure to determine and maintain the proper thumb contact pressure on the phone using combined guidance from the blood volume oscillations via the front camera and the adjacent screen touch sensor. Secondly, we eliminated the need for a force sensor by using the screen touch sensor as a surrogate and focusing on PP. We also showed that PP plus other readily available information can effectively detect systolic hypertension with ROC AUC of 0.9 by training and testing neural networks on a population-level database (see Figs. 3C, D). (PP with the other information could detect general hypertension (systolic BP ≥ 130 mmHg or diastolic BP ≥ 80 mmHg) with ROC AUC of 0.81 (not shown).)

While we came up with the main ideas prior to embarking on the study, it turned out to be a long journey to reach the final app and results. We ended up developing a number of different devices, as shown in Fig. 4, and studying the devices in volunteers with arm cuff PP as reference. Overall, we performed around 100 human studies under IRB approval in this work. We describe these initial studies below to importantly explain why the final app is the way it is.

Devices that were initially developed and studied in volunteers to arrive at the final app. (A) ‘Ideal sensors’ to implement the oscillometric hand raising/lowering method for measuring PP (Supplementary Information 2). (B) Wrist device to implement the PP measurement method (Supplementary Information 3). IR is infrared. (C) iPhone X app to implement the smartphone PP concept via ‘ideal guidance’ with a direct contact force measurement (Supplementary Information 4). (D) System to investigate screen touch parameters as a surrogate of contact pressure (Supplementary Information 5). (E) First version of an Android app for a conventional smartphone without force sensing involving an intended-to-be infrequent initialization step to determine the target thumb contact area followed by a measurement step (Supplementary Information 6). (F) The second version of the smartphone Android app involving hand lowering actuation (Supplementary Information 7). The subject in the figure provided informed consent to publish the images.

Publicité

We started with ‘ideal sensors’ to see if the oscillometric hand raising method even works (Fig. 4A, Supp. Mat. 3). We were concerned about the possibility of changes in smooth muscle contraction when raising the entire arm, which would violate the oscillometry assumption that the sigmoidal arterial blood volume-transmural pressure relationship is invariant during the measurement. Using a transmissive-mode infrared PPG clip with a Velcro tightening wrap and a fluid-filled tube-manometer system to directly measure ρgh, we found that PP could be measured fairly well via slow, continuous (rather than incremental) hand raising or lowering. This initial study convinced us that the oscillometric hand raising/lowering method is worthy of pursuit.

We studied wrist devices as an alternate way to implement the method (Fig. 4B, Supp. Mat. 4). While such devices are far less used than smartphones, the PP measurement would be more convenient in that the user would not have to maintain the contact pressure on the sensor. We found that devices embedded with PPG and direct ρgh sensors could measure PP well when the PPG sensor was placed over the radial artery but not the back of the wrist (similar to many consumer devices). However, locating the radial artery was too difficult. We concluded that smartwatches and fitness trackers currently in use cannot be converted into absolute BP sensors.

We began our study of smartphones using an iPhone X (Fig. 4C, Supp. Mat. 5). We could thus assess the feasibility of the user in maintaining the contact pressure during the hand raising via ‘ideal guidance’ with a direct contact force measurement (3D touch). We also initially explored the rear camera with flash for best measurement of the PPG waveform (from the index fingertip). However, smartphones have different rear camera configurations, so the app would have to be substantially modified for each phone. The front camera, on the other hand, is relatively standard and became our focus for greater generality. We found that users could indeed maintain the contact force during hand raising and lowering. We also determined that visual feedback of thumb contact was necessary, as there is a general tendency for the user to increase/decrease thumb contact pressure when performing hand raising/lowering due to the weight of the phone. We importantly found that holding the phone firmly with the supporting hand could mitigate this tendency.

We then studied the screen touch sensor as a surrogate for a force sensor (Fig. 4D, Supp. Mat. 6). We attached a force sensor to the back of a Samsung Galaxy S21 smartphone to measure the screen touch parameters (major and minor radii and x- and y-centroids) and thumb contact pressure simultaneously during thumb pressing. As expected, we found that the function relating each screen touch parameter to the thumb contact pressure increased progressively and then plateaued. However, we importantly discovered that the x-centroid plateaued last (i.e., it exhibited the greatest sensitivity at higher thumb contact pressures) and was least spurious of the touch parameters. We concluded that the touch x-centroid represents the thumb contact area best and could potentially guide the determination and maintenance of the thumb contact pressure.

We thereafter developed the first version of an Android app for the Samsung Galaxy S21 smartphone (Fig. 4E, Supp. Mat. 7). The app uses the front camera to measure the thumb PPG waveform, the z-axis accelerometer channel to measure ρgh, and the touch x-centroid sensor to measure the thumb contact area. In an ‘initialization step’, the user holds the device at heart level and slowly presses their thumb to increase and then decrease the blood volume oscillations while guided by the real-time and linear target thumb contact areas. The thumb contact area at the maximum oscillation amplitude (which occurs around the mean thumb BP) is selected as the target area. Then, in a ‘measurement step’, the user presses their thumb to reach this target area with the phone fully lowered and raises their hands in continuous motion over 20–40 s while maintaining the thumb contact area as guided additionally by a timer. The idea was that the initialization step need only be repeated when complete oscillograms are not obtained as a result of large BP changes in a user. The app also displays multiple rectangular boxes to determine the thumb position on the front camera and screen that yields the largest maximum oscillation for each user. The app could measure PP reasonably well via the same rectangular box for a number of users. However, it was too difficult to use. Firstly, linear thumb pressing was not easy, as the thumb contact area for guidance is not indicative of thumb contact pressure at high pressures. Secondly, the target thumb contact area was often not suitable due to variations in thumb positioning and pressing angle per measurement. We concluded that the target thumb contact area must be determined for each PP measurement and that the blood volume oscillation amplitude, which is determined by the thumb contact pressure when the device is at a fixed vertical level, should be used as a guide.

We thus developed a second version of the Android smartphone app (Fig. 4F, Supp. Mat. 8). The user holds the phone above their head and presses their thumb on the front camera and screen until the blood volume oscillations disappear. The app then displays the current thumb contact area as the target. The user thereafter lowers the phone in continuous motion while maintaining the thumb contact area on the target. Volunteers could achieve the arterial occlusion. However, very narrow oscillograms often resulted despite apparent maintenance of the thumb contact area over time. Based on our earlier studies, we reasoned that the thumb contact pressure must have decreased during hand lowering (which would increase the transmural pressure prematurely and narrow the oscillogram) but was not picked up by the thumb contact area measurement. We believe that thumb tissue viscoelasticity was the culprit (i.e., thumb contact area decreases slowly in response to a step decrease in thumb contact pressure). We concluded that the screen touch sensor is only a useful guide for hand raising in which the tendency is to increase the contact pressure.

We developed a third and final Android smartphone app by revisiting hand raising (see Fig. 2). The key challenge was to determine the target thumb contact area when the hands are in the initial fully lowered position. The reason is that confounding blood volume oscillations appear at low contact pressure due to arterioles and other smaller vessels in the thumb (see Fig. 5)¹⁸. So, simply having the user press their thumb on the phone until oscillations appear may often yield a target area that is too small to abolish the oscillations when the hands are fully raised. We ended up conceiving a heuristic procedure to overcome this challenge. The user presses on the phone until blood volume oscillations appear. The phone automatically detects the oscillations and then displays a target thumb contact area that is a fixed percentage higher than the current area. The user further presses until the current thumb contact area reaches the target. The user then slowly raises their hands while maintaining the thumb contact area on the target. The app computes PP from the function relating the blood volume oscillation amplitude via the front camera to ρgh via the z-axis accelerometer channel and the arm length (‘shifted oscillogram’). While skin tone is not a major concern for the app due to PPG measurement from the palmar side of the thumb where melanin is less abundant, cold environments do compromise visible-light PPG waveform quality.

PP computation algorithm of the smartphone app. The beats of the PPG waveform are detected. The peak-to-peak PPG amplitudes (PA), normalized to a maximum of unity, are plotted against the corresponding ρgh values to produce a discrete oscillogram. The oscillogram is smoothed and fitted to a function with adjustable parameters (\({A}_{n}\) and \({B}_{n}\)) to yield a continuous oscillogram. The derivative of this oscillogram is taken, and PP is detected as the width between its maximum and minimum¹⁸. HR is also determined from the beat detections.

We tested the app for usability in 24 participants (see Fig. 3A). We first explained and demonstrated how to use the app. An accompanying video tutorial could alternatively be included (Supp. Mat. 2). The procedures to be learned, which are not guided by the app, are holding the phone tightly with the supporting hand, pressing the thumb straight down, and pressing at a consistent pace to transition from a noisy signal to blood volume oscillations, and locking the wrist-hand angle during the hand raise. Of these procedures, the steepest learning curve was pressing the thumb to elicit blood volume oscillations at relatively low thumb contact. Yet, after 6–7 practice trials, new users were able to make valid measurements with 60% success rate. Most users could start making valid measurements after 3–4 practice trials. The success rate reached 80% for experienced users and may increase further with regular app usage. We hypothesize that regular app usage may also minimize any changes in the user’s PP that could potentially occur due to the act of performing the thumb and hand maneuvers, which could increase stress for example.

The heuristic procedure for determining the target thumb contact area was effective in 22 of the 24 participants. In the remaining two participants, the oscillograms did not show an ascending limb despite proper app usage. The fixed percentage increase was simply lowered by half with the app, and complete oscillograms were then readily obtained. Conversely, if the oscillogram does not show a descending limb, then the fixed percentage increase could be raised.

We also tested the app for accuracy against a validated automatic arm cuff device in the same participants (see Fig. 3B). The PP via the app yielded r = 0.70 and an overall error 3), thereby further indicating that the users were able to effectively perform the thumb and hand maneuvers.

There are several sources of error. Firstly, we used the classic derivative algorithm to compute PP. This algorithm can amplify noise and underestimate PP in the thumb artery, especially in the higher PP range, due to small artery viscoelasticity (see Fig. 3B)²³. However, it is possible to overcome viscoelasticity, while avoiding derivatives, to accurately compute PP²³. Secondly, the reference device that we used measures arm PP. Thumb PP may be variably higher than arm PP due to arterial wave reflection²⁴. We did employ a volume-clamp cuff device to obtain a finger BP waveform. However, as we previously found¹⁸, the thumb PP correlated better with the more important arm cuff PP. Thirdly, we used a less reliable automatic cuff device as reference.

Another limitation of our study was that few participants with Stage 2 hypertension were included. However, we expect that our smartphone PP app can be effective in hypertension. In our studies of the Android app, we observed that the effective thumb contact area range was about 75–95% of the maximum thumb contact area of the users and that the target thumb contact area corresponded on average to 85% of the maximum. Based on an exponential function for relating touch x-centroid to the thumb contact pressure with average parameter values (Supp. Mat. 6), a target thumb contact area of 90% of the maximum would correspond to a thumb contact pressure of 125 mmHg. Since mean thumb BP is about 10 mmHg lower than mean arm BP²⁴, the app may thus be applicable in hypertensive users with mean BP of around 135 mmHg (e.g., 180/110 mmHg). Although the app may not be effective in users with higher mean BP or, since the BP swing via hand raising is typically about 80 mmHg, very wide PP (e.g., 120 mmHg), it is designed for people 

The major limitation of the smartphone PP app is indeed that it is not applicable to much of the elderly population. It is well known that the elderly are most afflicted by systolic hypertension. However, while regular BP measurement in the elderly is important to help with hypertension control, detection of hypertension in this population may not be vital. The reason is that virtually every person older than ~ 70 years may be sensibly assumed to be at a systolic BP level of increased risk for cardiovascular events²⁵. So, from a hypertension detection perspective, the younger and middle-aged populations may actually be most relevant.

Future research is needed to bring the smartphone PP app to practice. Firstly, an accurate PP algorithm must be developed. We suggest to collect training data comprising ideal sensor measurements and manual auscultatory cuff BP measurements as reference from a large cohort of participants of diverse BP levels. The assumption here is that user error is random and could not be compensated for algorithmically. Secondly, this app must be tested rigorously for accuracy and usability in the field. Thirdly, the app should be tested with various smartphones due to differences in functionality (e.g., maximal screen brightness). It may also be worthwhile to explore obtaining diastolic BP information with the app. For example, if the target thumb contact area is low relative to the maximum thumb contact area of the user, then it may be reasonable to conclude low diastolic BP. That said, it is high systolic BP that is the most important modifiable risk factor²⁶.

Following successful future research, the smartphone PP app (Android, iOS, and other versions) could be used by anyone who has or knows someone with a smartphone and connectivity to download the app. The app may be especially of interest to young and middle-aged adults in underserved populations who are enthusiastic about maintaining their health but have limited or no access to BP cuff devices. Such users could readily take action upon learning that their usual PP is high through regular app usage. For example, even if procuring medication is difficult, they could still mitigate their condition by eating a healthy diet, maintaining normal body weight, increasing physical activity, and avoiding excessive alcohol²⁷. Ultimately, the smartphone PP app could help reduce the massive burden of systolic hypertension in underserved populations and thus global health inequalities.